Date: Sat, 5 Jan 2002 21:46:44 -0800
From: Kamal A Mostafa <kamal@whence.com>
To: Riccardo Facchetti <riccardo@master.oasi.gpa.it>, Kern Sibbald <kern@sibbald.com>
Subject: new EVENTSFILEMAX code

Hi Guys-

Well, I went ahead and rewrote truncate_events_file() from scratch -- it was
a nice file-size-based thing that used no malloc's.  Alas, while testing it I
discovered a rather nasty bug which, I believe, has been present since at
least 3.8.4...

I was running some tests where I'd fill up the events file to force the new
truncation code to kick in.  The new code worked fine, but when I killed
off apcupsd, I noticed that I didn't get the usual "apcupsd exiting, signal
15" and "apcupsd shutdown succeeded" log_event() messages that should have
been at the tail end of the eventsfile.

I suspect that the function terminate() is using an old copy of the event_fd
file descriptor (which has since been closed) or something along those lines.
As I see it, any truncation scheme which closes and reopens the ups->evend_fd
would cause this problem.  Maybe truncation screws up the forked children or
threads also.  Bummer.

So I rewrote it again.  My new trim_eventfile() doesn't ever close the
ups->event_fd -- that's probably just "better" anyway.  The basic premise is
rewind, read, rewind, truncate, write.  And sure enough, now I get the
"exiting" messages even after the file has been trimmed, so I think all is
well.

*Whew*  I'm sorry I ever started poking at it in the first place!  ;-}


Anyway...  The new version replaces the TRIMEVENTSFILE [on|off] param with
"EVENTSFILEMAX <kilobytes>".  I've attached two versions of the patch.  Use
one or the other, depending on whether you have or haven't already applied
my previous TRIMEVENTSFILE patch.

        apc.evmax-398.patch     -applies to apcupsd-3.9.8-18Dec01

        apc.evmax-kamal.patch   -applies to TRIMEVENTSFILE version

Comments?

===================

KES: Integrated 9Jan02 there were a number of rejects due to space/tab
     mangling.